Precision
Cooking.
How It Works
Multi-agent system that routes, retrieves, evaluates, and remembers.
Router
Determines if the question is cookbook-related, a catalog request, or out-of-scope. Handles follow-ups intelligently.
Retrieval
Searches the cookbook using semantic vector search (Chroma). Returns relevant recipe chunks with metadata (source, page).
LLM
Synthesizes an answer using ONLY the retrieved cookbook context. Supports streaming tokens for real-time UX.
Evaluation
Verifies the answer is supported by retrieved chunks. Produces confidence scores and facts-checked metadata.
Memory
Persists chat messages and recipe context in SQLite. Enables follow-ups and maintains session history.
Response
Returns the final answer with sources (full book name + page), evaluation metrics, and structured metadata.
Ask RecipaAI
RecipaAI may make mistakes.
